You bring the hull. We build the layer that makes it a vessel.

The six metre uncrewed hull under way in a short sea, bow lifted, throwing spray down its port side.

Sensor to mission

An operator drags a box on live video and the vessel has a waypoint.

The gimbal locks and ray casts to the surface. The MISB ST 0601 target position is demuxed from the video stream and dropped as a mission item. The vessel drives to it.

QGroundControl has gimbal control and no KLV decoder. Mission Planner has neither. Neither can turn a video click into a vehicle action.

Fig. 1 Side elevation. The red mark is the operator: the box drawn on the video, and the waypoint it becomes. Everything else is the vessel doing what it is told.

Any sensor.
Any propulsion.

The two and a half metre hull seen from abeam, low in the water, drive at the transom.

2.5 mThe smaller hull

What Lanchester integrates, and how long a new integration takes
Subsystem	Supported	Detail
Sensors	Any	Electro-optical, infrared, gimbals, radar, AIS, sonar. If it emits data, we will take it.
Propulsion	Any	Diesel, outboard, electric pod, waterjet. If it accepts a command, we will drive it.
New integration	Days	Not months. The integration layer was built first, so a new payload is an adapter rather than a new system.
Custom builds	On request	Tell us the hull, the sensor and the drive. We will tell you what the integration looks like.
Autopilot	ArduPilot	Open source, run as shipped. Mature, widely reviewed, and deliberately not our maintenance burden. Station keeping and the stock hold modes come from it, not from us.

Not knowing is
a third answer

Every depth verdict is safe, unsafe, or unknown.

A two-state gate has to call unsurveyed water either safe or unsafe, and both answers are lies. Unknown never blocks a waypoint an operator places by hand, because the absence of a charted shoal is not evidence of one. It does refuse a generated route: proposing a passage across water nothing has verified is a different act.

Every refusal names the chart cell and the record that produced it, and exports as a record rather than a screenshot.

>>> check(29.9330, -90.0600)    Mississippi, mid-channel
SAFE      10.0 m charted, 2.0 m required (1.5 m draft + 0.5 m margin)

>>> check(36.6595, -121.9563)   a charted wreck off Monterey
UNSAFE    charted wreck 0 m away (US3CA1FF/396), depth over it NOT GIVEN

>>> check(29.0000, -95.5000)    Gulf of Mexico, off-folio
UNKNOWN   no charted depth area covers this position

Run against the chart folio in the repository. The draft comes from the hull’s own profile, not a number typed in by hand. No charted depth is not clear water, and a gate that says so is worth more than one that guesses.

The operator console over the Mississippi at New Orleans. A generated five leg route is loaded as the plan, fifteen AIS contacts are tracked and named, the header warns that the mission crosses charted shallow, and the console is labelled SIMULATOR, NOT A VESSEL. — Fig. 3 The same gate in the console, over the Mississippi at New Orleans. A route has been generated to the chart, five legs and 6.40 km, and the header states that it crosses charted shallow before anything runs. Fifteen AIS contacts tracked and named, chart edition and scale on the face, every tile served from disk. The badge reads **simulator, not a vessel**, because it is one.

Demonstrated,
and not

The six metre hull under way with its communications mast stepped amidships.

6 mThe larger hull

On the waterDemonstrated

A vessel in Australia, commanded from a desk in California, over a link we did not own.
Autonomous waypoint navigation, in sea trials.
Operation under degraded communications.
Offline console, charts pre-cached, 291,087 tiles served from disk. No third-party tile host is reachable from the shipped console.
Multi-link command and control, with bandwidth shaped per link.

In simulationNot yet on hardware

Positioning that holds when GNSS does not.
Cooperative behaviour across more than one vessel.
Closed-loop validated. The sensors are not fitted, so none of it is claimed on the water.

We draw these lines ourselves so nobody has to draw them for us.

Cross-track error, speed profile and turn performance were measured on the water over three days in August 2026. Every figure is from MAVLink telemetry rather than a datasheet, and nothing is quoted that was not measured: no power module was fitted, so we hold no energy or efficiency numbers and publish none. Specification sheets on request.

Who we work with

Shipyards and boat builders with hulls and no autonomy software. Integrators assembling a USV offering from several vendors. Primes that need a command and control layer not tied to one platform or one sensor.

Contact

jesse@lanchester.ai

Custom builds are welcome. You only have to ask.